* Perf: stream manifest parsing with ijson in document_importer
Replace bulk json.load of the full manifest (which materializes the
entire JSON array into memory) with incremental ijson streaming.
Eliminates self.manifest entirely — records are never all in memory
at once.
- Add ijson>=3.2 dependency
- New module-level iter_manifest_records() generator
- load_manifest_files() collects paths only; no parsing at load time
- check_manifest_validity() streams without accumulating records
- decrypt_secret_fields() streams each manifest to a .decrypted.json
temp file record-by-record; temp files cleaned up after file copy
- _import_files_from_manifest() collects only document records (small
fraction of manifest) for the tqdm progress bar
Measured on 200 docs + 200 CustomFieldInstances:
- Streaming validation: peak memory 3081 KiB -> 333 KiB (89% reduction)
- Stream-decrypt to file: peak memory 3081 KiB -> 549 KiB (82% reduction)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Perf: slim dict in _import_files_from_manifest, discard fields
When collecting document records for the file-copy step, extract only
the 4 keys the loop actually uses (pk + 3 exported filename keys) and
discard the full fields dict (content, checksum, tags, etc.).
Peak memory for the document-record list: 939 KiB -> 375 KiB (60% reduction).
Wall time unchanged.
* Refactor: migrate exporter/importer from tqdm to PaperlessCommand.track()
Replace direct tqdm usage in document_exporter and document_importer with
the PaperlessCommand base class and its track() method, which is backed by
Rich and handles --no-progress-bar automatically. Also removes the unused
ProgressBarMixin from mixins.py.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Refactor: add explicit supports_progress_bar and supports_multiprocessing to all PaperlessCommand subclasses
Each management command now explicitly declares both class attributes
rather than relying on defaults, making intent unambiguous at a glance.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Perf: streaming manifest writer for document exporter (Phase 3)
Replaces the in-memory manifest dict accumulation with a
StreamingManifestWriter that writes records to manifest.json
incrementally, keeping only one batch resident in memory at a time.
Key changes:
- Add StreamingManifestWriter: writes to .tmp atomically, BLAKE2b
compare for --compare-json, discard() on exception
- Add _encrypt_record_inline(): per-record encryption replacing the
bulk encrypt_secret_fields() call; crypto setup moved before streaming
- Add _write_split_manifest(): extracted per-document manifest writing
- Refactor dump(): non-doc records streamed during transaction, documents
accumulated then written after filenames are assigned
- Upgrade check_and_write_json() from MD5 to BLAKE2b
- Remove encrypt_secret_fields() and unused itertools.chain import
- Add profiling marker to pyproject.toml
Measured improvement (200 docs + 200 CustomFieldInstances, same
dump() code path, only writer differs):
- Peak memory: ~50% reduction
- Memory delta: ~70% reduction
- Wall time and query count: unchanged
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Refactor: O(1) lookup table for CRYPT_FIELDS in per-record encryption
Add CRYPT_FIELDS_BY_MODEL to CryptMixin, derived from CRYPT_FIELDS at
class definition time. _encrypt_record_inline() now does a single dict
lookup instead of a linear scan per record, eliminating the loop and
break pattern.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Phase 1 -- Eliminate JSON round-trip in document exporter
Replace json.loads(serializers.serialize("json", qs)) with
serializers.serialize("python", qs) to skip the intermediate
JSON string allocation and parse step. Use DjangoJSONEncoder
in check_and_write_json() to handle native Python types
(datetime, Decimal, UUID) the Python serializer returns.
Phase 2 -- Batched QuerySet serialization in document exporter
Add serialize_queryset_batched() helper that uses QuerySet.iterator()
and itertools.islice to stream records in configurable chunks, bounding
peak memory during serialization to batch_size * avg_record_size rather
than loading the entire QuerySet at once.
